We present an end-to-end decision-support pipeline that automates the daily pricing and replenishment of fresh vegetables in a supermarket while respecting cost-plus rules, shelf-space limits, and spoilage. First, three years of point-of-sale data (251 SKUs, six categories) are explored: seasonality, category complementarity, and near-normal demand distributions are confirmed via descriptive statistics, Pearson correlation, and Kolmogorov–Smirnov tests. Grey prediction GM(2,1) models then generate one-week wholesale-cost forecasts whose mean absolute error is under 11%. Next, ordinary least-squares fits deliver category-level price–demand elasticities ( \(R^{2}=0.59\!-\!0.88\) ), which, together with the grey costs, feed a non-linear profit-maximisation programme. Simulated annealing solves this joint price-and-quantity problem in <0.5s and lifts projected weekly profit by 12–15% relative to a static mid-band policy. Finally, a 0–1 model selects the optimal 30 SKUs from the daily candidate pool of 50, using implicit enumeration to respect shelf capacity and face-up minima; the resulting mix raises expected margin by a further 6% while reducing shelf congestion by 40%. Back-testing on July 2022 data shows that predicted profits deviate from actual outcomes by less than 10% on most days. The methodology combining gray forecasting, linear elasticity estimation, metaheuristic nonlinear programming, and binary space optimization, forms a coherent, data-driven framework that can be generalized to other perishable-goods contexts.